Closed
Description
Initial Checks
- I confirm that I'm using the latest version of Pydantic AI
- I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue
Description
Not a bug, just a learning to share in case others make the same mistake!
Recently, I had a few infinite loop agents despite having set usage_limits -- after some debugging, I realized that I was passing usage_limits key to the Agent, eg, Agent(..., usage_limits=UsageLimits(request_limit=3)).
I had a lot of run() calls, so I suspect I wanted to set it globally.
The correct usage based on the docs is to pass it to each invocation of the run*, eg, run(..., usage_limits=UsageLimits(request_limit=3))
My "wish" would be that providing an unsupported kwarg to the Agent would fail explicitly -- not sure where this unsupported argument got slotted.
See a modification of the docs example that DOES NOT WORK:
Example Code
# modification of https://ai.pydantic.dev/agents/#usage-limits
from typing_extensions import TypedDict
from pydantic_ai import Agent, ModelRetry
from pydantic_ai.exceptions import UsageLimitExceeded
from pydantic_ai.usage import UsageLimits
class NeverOutputType(TypedDict):
"""
Never ever coerce data to this type.
"""
never_use_this: str
agent = Agent(
'openai:gpt-4.1-mini',
retries=3,
output_type=NeverOutputType,
system_prompt='Any time you get a response, call the `infinite_retry_tool` to produce another response.',
# THIS IS WRONG!!! IT WILL NOT WORK but it fails silently
usage_limits=UsageLimits(request_limit=3)
)
counter = 0
@agent.tool_plain(retries=5)
def infinite_retry_tool() -> int:
global counter
counter += 1
print(f"Called infinite_retry_tool {counter} times")
if counter == 5:
raise Exception("Too many retries")
raise ModelRetry('Please try again.')
try:
result_sync = agent.run_sync(
'Begin infinite retry loop!'
)
except UsageLimitExceeded as e:
print(e)
#> The next request would exceed the request_limit of 3Python, Pydantic AI & LLM client version
0.2.18